The objectives (s) of this project is (are): One mechanism for achieving improved performance in distributed computing systems is dynamic process migration, the movement of processes from one node to another in response to dynamic changes in system loads or due to the dynamic behavior of distributed computations. The goal of this research is to investigate ways to use knowledge about process characteristics in the design of migration algorithms. An empirical study of existing distributed computations will be performed to determine what kinds of process traits are important for migration decisions, the difficulty and cost of ascertaining these traits, and the relation between static predictions of process behavior and the actual dynamic characteristics exhibited at execution time are being design. Migration algorithms that use knowledge about these process traits (1) to decide which process to migrate and (2) to ok migration based on both static and dynamic interprocess communication patterns. Both graph based/network flow type algorithms as well as decentralized negotiation type algorithms will be used. These algorithms will be tested through simulation on a variety of distributed computations and under varying system loads. This work will help distributed systems reach the full power of compute sharing by enriching our understanding of distributed computation and through the insights into policies and mechanisms needed to guide and provide process migration.
